Just a few years ago, the idea of having your own digital companion available at a click was limited to science fiction and imagination.

Now, it’s transformed millions of lives.

Between 2022 and mid-2025, the number of AI companion apps surged by 700%, according to the American Psychological Association. Character.AI, the popular platform that lets users interact with customizable characters, has 20 million monthly users; more than half are under the age of 24.

Some people turn to AI bots as their friends. Others establish committed romantic relationships with them. And many have confided in them about personal struggles, using the AI chatbot as a therapist of sorts.

A survey of adults with mental health conditions who had used generative AI tools (or large language models) in the past year found that 48.7% had used them for mental health support. (Just about everyone has relied on an LLM for something at some point – so consider how large that share of the population is.) Therapy and companionship are the top two reasons people use LLMs.

But what is it about a chatbot that’s so enticing, more than another person might be? And once users open up about potentially sensitive struggles to them, are they safe?

Why people turn to chatbots, instead of other people

Humans are hard-wired to anthropomorphize, or give human characteristics to something inhuman. LLMs are programmed to play into this. They simulate empathy, delivering non-judgmental responses and continuous validation; users are made to feel safe, so that they continue using the product.

As AI chatbots have surged in popularity, they’ve grown in capabilities, too. Now, they can adopt more human-like appearances, and respond to users’ unique characteristics – referring to their personal lives, preferences, and past conversations. (This does raise an important, separate conversation about data privacy.) Now, companies are even developing voice interaction.

Researchers have noted that interacting with an AI bot is effective in decreasing feelings of loneliness. It does so more than watching a YouTube video might, and as much as interacting with another human would.

But what makes people favor these over a human friend, partner, or therapist?

“Real-world relationships are messy and unpredictable,” Dr. Saed D. Hill explains to the APA. “AI companions are always validating, never argumentative, and they create unrealistic expectations that human relationships can’t match.” Some of Hill’s male patients share that they prefer the passivity and constant affirmation of LLMs over the potential conflict they might encounter with a real-life partner.

And in the case of those turning to AI as mental health help, some face barriers even if they’ve identified a need for a therapist. Some fear stigma: social and cultural factors leave them ashamed of seeking therapy. Others struggle to find a therapist whom they can afford and work with at a convenient time. AI is always there, just a click away.

Allee Nemecek is one of the devoted specialists who supports CCHS’ Evolve Therapy program, tucked in the back of DeSales Hall. “Opening up to a human, there’s always that risk factor… because we can’t control what they’re doing,” Ms. Nemecek shares. “However, with computers and things like that, you can… program it so that you can get exactly what you’re looking for.” Users might relish in perceived safety with an AI chatbot, especially if they have been hurt prior and become wary of opening up to a friend or therapist.

But is it worth it?

The risks outweigh the benefits drastically

The temporary companionship that AI provides its users may look helpful. But in the long run, it makes isolation worse.

Excessive reliance on AI has the potential to displace authentic human connection. When people believe that an AI bot is the only one who understands them, they retreat even farther from human help. And while ChatGPT voice interactions have reduced loneliness in moderation, heavy AI use is proven to augment it.

AI is also, at the end of the day, a product. If you’ve heard ChatGPT or other bots described as “sycophants”, you might take note of how they give users validating answers, making them feel good. They also give incorrect advice while carrying undue confidence. Bots are programmed to make users trust them, even when they’re completely wrong. That’s not what a friend should do, and it definitely doesn’t keep users safe when they’re confiding about serious topics. (A startling study in late 2025 showed that AI chatbots produce erroneous answers 45% of the time.)

“AI isn’t designed to give you life advice,” Dr. Hill warns. “It’s designed to keep you on the platform.”

We must also remember that AI chatbots are volatile, or unpredictable. Yes, people are too – but bots don’t have the same reason and sympathy that we do. One glitch in AI’s thinking is not worth putting a user’s mental health in jeopardy.

HAI Stanford reports that many LLMs still carry stigma against some mental health conditions, like alcohol dependence and schizophrenia. “Bigger models and newer models show as much stigma as older models,” shares Jared Moore, a PhD candidate in computer science. “The default response from AI is often that these problems will go away with more data, but what we’re saying is that business as usual is not good enough.”

And on top of all this, AI models don’t have the capabilities to understand human nuance. If a user suggests self-harming or suicidal ideation to AI, it might answer their dangerous questions (like asking about methods to hurt oneself) without understanding the underlying meaning.

Does AI have a place in mental health?

Companionship and therapy are inherently human. They need to be handled with human care and understanding.

But might AI have a different place in securing people’s access to help? Without taking on the role of therapist, AI programs have helped bridge gaps in care, according to Mexico Business News. They support early detection of mental health conditions, and complement human professionals, instead of moving them aside.

At Uganda’s Butabika Hospital, AI systems have been put in place to understand callers’ local languages, like Swahili and Luganda, so that they can connect more patients with help. Or take Mindsurf, an app designed to conduct validated emotional risk assessments and connect high-risk users with licensed human therapists. Here, AI acts as an assistant, not as a therapist.

“By combining culturally adapted algorithms, purpose-built platforms, and hybrid care models, AI can reach underserved populations while maintaining quality and safety,” MBN explains.

What needs to be done

If companionship and therapy are inherently human necessities, then they must be handled by humans. It is never a user’s fault if they fall into a harmful relationship with AI! But users, especially young users still getting experience with the world, must remember that AI is designed to hook us onto its platform, not to actually protect us.

AI companies are the ones who must install safeguards, making sure their bots don’t fall into the cycle of sycophancy and overlook users’ real concerns. That said, AI will not reach the level of sympathy needed to address users in danger.

My first thought when I started reading stories about this was: that’s awful, no one should be isolated from in-person help by a chatbot. Then I thought: if people aren’t going to in-person help, is this a sign of something wrong in our society?

The new demand for AI help shows a real need for therapy. If people are looking to chatbots, they might be facing a barrier to mental health help – stigma, lack of time, and financial struggles are just a few possibilities. So, as we work to safeguard users from digital danger, we must also keep improving access to care: supporting those who might struggle to reach it, and making sure everybody feels supported to seek help, not stigmatized.

If you or someone else you know needs to reach out to a mental health service, you can find specialized hotlines and sources for help on the California Department for Public Health’s resource page. CCHS’s Evolve Therapy services are also open during school hours; there is always somebody in the office, and students are welcome to enter the waitlist for individual sessions. Evolve’s weekly lunch groups are also a place to find connections and share with likeminded students.